About

Rosso Cyber Cafe is the publishing home for Angelo Rosso’s independent work on AI agent reliability. It is one person, working in the open, without a company behind it.

The premise is that agent systems are being wired into things that move money and change records, and that the industry is still evaluating them mostly on whether they produce good text. That is the wrong axis. What matters is what an agent does when a tool call times out, a response is lost, a retry fires, or a process dies half-way through a write. So the work here is deliberately adversarial: construct the failure, run it enough times to get a number, and publish the number whether or not it is flattering. Every note should contain enough detail that you can reproduce it, and enough honesty that you can tell me where it is wrong.

The tools are a by-product. When an experiment needs a harness, the harness gets cleaned up and released — once is the first. All of it is on GitHub, MIT licensed, with no upsell attached.

Contact

Corrections and disagreements are genuinely welcome — especially if you have run the same experiment and got a different result.